1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "compiler/compiler_globals.hpp"
  26 #include "interp_masm_x86.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "logging/log.hpp"
  30 #include "oops/arrayOop.hpp"
  31 #include "oops/markWord.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/method.hpp"
  34 #include "oops/resolvedFieldEntry.hpp"
  35 #include "oops/resolvedIndyEntry.hpp"
  36 #include "oops/resolvedMethodEntry.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/basicLock.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/javaThread.hpp"
  42 #include "runtime/safepointMechanism.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "utilities/powerOfTwo.hpp"
  45 
  46 // Implementation of InterpreterMacroAssembler
  47 
  48 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  49   assert(entry, "Entry must have been generated by now");
  50   jump(RuntimeAddress(entry));
  51 }
  52 
  53 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
  54   Label update, next, none;
  55 
  56   assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
  57 
  58   interp_verify_oop(obj, atos);
  59 
  60   testptr(obj, obj);
  61   jccb(Assembler::notZero, update);
  62   testptr(mdo_addr, TypeEntries::null_seen);
  63   jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
  64   // atomic update to prevent overwriting Klass* with 0
  65   lock();
  66   orptr(mdo_addr, TypeEntries::null_seen);
  67   jmpb(next);
  68 
  69   bind(update);
  70   load_klass(obj, obj, rscratch1);
  71   mov(rscratch1, obj);
  72 
  73   xorptr(obj, mdo_addr);
  74   testptr(obj, TypeEntries::type_klass_mask);
  75   jccb(Assembler::zero, next); // klass seen before, nothing to
  76                                // do. The unknown bit may have been
  77                                // set already but no need to check.
  78 
  79   testptr(obj, TypeEntries::type_unknown);
  80   jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
  81 
  82   cmpptr(mdo_addr, 0);
  83   jccb(Assembler::equal, none);
  84   cmpptr(mdo_addr, TypeEntries::null_seen);
  85   jccb(Assembler::equal, none);
  86 
  87   // There is a chance that the checks above (re-reading profiling
  88   // data from memory) fail if another thread has just set the
  89   // profiling to this obj's klass
  90   mov(obj, rscratch1);
  91   xorptr(obj, mdo_addr);
  92   testptr(obj, TypeEntries::type_klass_mask);
  93   jccb(Assembler::zero, next);
  94 
  95   // different than before. Cannot keep accurate profile.
  96   orptr(mdo_addr, TypeEntries::type_unknown);
  97   jmpb(next);
  98 
  99   bind(none);
 100   // first time here. Set profile type.
 101   movptr(mdo_addr, obj);
 102 #ifdef ASSERT
 103   andptr(obj, TypeEntries::type_klass_mask);
 104   verify_klass_ptr(obj);
 105 #endif
 106 
 107   bind(next);
 108 }
 109 
 110 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
 111   if (!ProfileInterpreter) {
 112     return;
 113   }
 114 
 115   if (MethodData::profile_arguments() || MethodData::profile_return()) {
 116     Label profile_continue;
 117 
 118     test_method_data_pointer(mdp, profile_continue);
 119 
 120     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
 121 
 122     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
 123     jcc(Assembler::notEqual, profile_continue);
 124 
 125     if (MethodData::profile_arguments()) {
 126       Label done;
 127       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
 128       addptr(mdp, off_to_args);
 129 
 130       for (int i = 0; i < TypeProfileArgsLimit; i++) {
 131         if (i > 0 || MethodData::profile_return()) {
 132           // If return value type is profiled we may have no argument to profile
 133           movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 134           subl(tmp, i*TypeStackSlotEntries::per_arg_count());
 135           cmpl(tmp, TypeStackSlotEntries::per_arg_count());
 136           jcc(Assembler::less, done);
 137         }
 138         movptr(tmp, Address(callee, Method::const_offset()));
 139         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
 140         // stack offset o (zero based) from the start of the argument
 141         // list, for n arguments translates into offset n - o - 1 from
 142         // the end of the argument list
 143         subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
 144         subl(tmp, 1);
 145         Address arg_addr = argument_address(tmp);
 146         movptr(tmp, arg_addr);
 147 
 148         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
 149         profile_obj_type(tmp, mdo_arg_addr);
 150 
 151         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
 152         addptr(mdp, to_add);
 153         off_to_args += to_add;
 154       }
 155 
 156       if (MethodData::profile_return()) {
 157         movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 158         subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
 159       }
 160 
 161       bind(done);
 162 
 163       if (MethodData::profile_return()) {
 164         // We're right after the type profile for the last
 165         // argument. tmp is the number of cells left in the
 166         // CallTypeData/VirtualCallTypeData to reach its end. Non null
 167         // if there's a return to profile.
 168         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
 169         shll(tmp, log2i_exact((int)DataLayout::cell_size));
 170         addptr(mdp, tmp);
 171       }
 172       movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
 173     } else {
 174       assert(MethodData::profile_return(), "either profile call args or call ret");
 175       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
 176     }
 177 
 178     // mdp points right after the end of the
 179     // CallTypeData/VirtualCallTypeData, right after the cells for the
 180     // return value type if there's one
 181 
 182     bind(profile_continue);
 183   }
 184 }
 185 
 186 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
 187   assert_different_registers(mdp, ret, tmp, _bcp_register);
 188   if (ProfileInterpreter && MethodData::profile_return()) {
 189     Label profile_continue;
 190 
 191     test_method_data_pointer(mdp, profile_continue);
 192 
 193     if (MethodData::profile_return_jsr292_only()) {
 194       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
 195 
 196       // If we don't profile all invoke bytecodes we must make sure
 197       // it's a bytecode we indeed profile. We can't go back to the
 198       // beginning of the ProfileData we intend to update to check its
 199       // type because we're right after it and we don't known its
 200       // length
 201       Label do_profile;
 202       cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
 203       jcc(Assembler::equal, do_profile);
 204       cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
 205       jcc(Assembler::equal, do_profile);
 206       get_method(tmp);
 207       cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
 208       jcc(Assembler::notEqual, profile_continue);
 209 
 210       bind(do_profile);
 211     }
 212 
 213     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
 214     mov(tmp, ret);
 215     profile_obj_type(tmp, mdo_ret_addr);
 216 
 217     bind(profile_continue);
 218   }
 219 }
 220 
 221 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
 222   if (ProfileInterpreter && MethodData::profile_parameters()) {
 223     Label profile_continue;
 224 
 225     test_method_data_pointer(mdp, profile_continue);
 226 
 227     // Load the offset of the area within the MDO used for
 228     // parameters. If it's negative we're not profiling any parameters
 229     movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
 230     testl(tmp1, tmp1);
 231     jcc(Assembler::negative, profile_continue);
 232 
 233     // Compute a pointer to the area for parameters from the offset
 234     // and move the pointer to the slot for the last
 235     // parameters. Collect profiling from last parameter down.
 236     // mdo start + parameters offset + array length - 1
 237     addptr(mdp, tmp1);
 238     movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
 239     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 240 
 241     Label loop;
 242     bind(loop);
 243 
 244     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
 245     int type_base = in_bytes(ParametersTypeData::type_offset(0));
 246     Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
 247     Address arg_off(mdp, tmp1, per_arg_scale, off_base);
 248     Address arg_type(mdp, tmp1, per_arg_scale, type_base);
 249 
 250     // load offset on the stack from the slot for this parameter
 251     movptr(tmp2, arg_off);
 252     negptr(tmp2);
 253     // read the parameter from the local area
 254     movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
 255 
 256     // profile the parameter
 257     profile_obj_type(tmp2, arg_type);
 258 
 259     // go to next parameter
 260     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 261     jcc(Assembler::positive, loop);
 262 
 263     bind(profile_continue);
 264   }
 265 }
 266 
 267 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
 268                                                   int number_of_arguments) {
 269   // interpreter specific
 270   //
 271   // Note: No need to save/restore bcp & locals registers
 272   //       since these are callee saved registers and no blocking/
 273   //       GC can happen in leaf calls.
 274   // Further Note: DO NOT save/restore bcp/locals. If a caller has
 275   // already saved them so that it can use rsi/rdi as temporaries
 276   // then a save/restore here will DESTROY the copy the caller
 277   // saved! There used to be a save_bcp() that only happened in
 278   // the ASSERT path (no restore_bcp). Which caused bizarre failures
 279   // when jvm built with ASSERTs.
 280 #ifdef ASSERT
 281   {
 282     Label L;
 283     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 284     jcc(Assembler::equal, L);
 285     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 286          " last_sp != null");
 287     bind(L);
 288   }
 289 #endif
 290   // super call
 291   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 292   // interpreter specific
 293   // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
 294   // but since they may not have been saved (and we don't want to
 295   // save them here (see note above) the assert is invalid.
 296 }
 297 
 298 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
 299                                              Register last_java_sp,
 300                                              address  entry_point,
 301                                              int      number_of_arguments,
 302                                              bool     check_exceptions) {
 303   // interpreter specific
 304   //
 305   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
 306   //       really make a difference for these runtime calls, since they are
 307   //       slow anyway. Btw., bcp must be saved/restored since it may change
 308   //       due to GC.
 309   save_bcp();
 310 #ifdef ASSERT
 311   {
 312     Label L;
 313     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 314     jcc(Assembler::equal, L);
 315     stop("InterpreterMacroAssembler::call_VM_base:"
 316          " last_sp isn't null");
 317     bind(L);
 318   }
 319 #endif /* ASSERT */
 320   // super call
 321   MacroAssembler::call_VM_base(oop_result, last_java_sp,
 322                                entry_point, number_of_arguments,
 323                                check_exceptions);
 324   // interpreter specific
 325   restore_bcp();
 326   restore_locals();
 327 }
 328 
 329 void InterpreterMacroAssembler::call_VM_preemptable_helper(Register oop_result,
 330                                                            address entry_point,
 331                                                            int number_of_arguments,
 332                                                            bool check_exceptions) {
 333   Label resume_pc, not_preempted;
 334 
 335 #ifdef ASSERT
 336   {
 337     Label L;
 338     cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
 339     jcc(Assembler::equal, L);
 340     stop("Should not have alternate return address set");
 341     bind(L);
 342   }
 343 #endif /* ASSERT */
 344 
 345   // Force freeze slow path.
 346   push_cont_fastpath();
 347 
 348   // Make VM call. In case of preemption set last_pc to the one we want to resume to.
 349   // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
 350   lea(rscratch1, resume_pc);
 351   push(rscratch1);
 352   MacroAssembler::call_VM_helper(oop_result, entry_point, number_of_arguments, check_exceptions);
 353   pop(rscratch1);
 354 
 355   pop_cont_fastpath();
 356 
 357   // Check if preempted.
 358   movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
 359   cmpptr(rscratch1, NULL_WORD);
 360   jccb(Assembler::zero, not_preempted);
 361   movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
 362   jmp(rscratch1);
 363 
 364   // In case of preemption, this is where we will resume once we finally acquire the monitor.
 365   bind(resume_pc);
 366   restore_after_resume(false /* is_native */);
 367 
 368   if (check_exceptions) {
 369     // check for pending exceptions (java_thread is set upon return)
 370     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 371     Label ok;
 372     jcc(Assembler::equal, ok);
 373     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 374     bind(ok);
 375   }
 376 
 377   // get oop result if there is one and reset the value in the thread
 378   if (oop_result->is_valid()) {
 379     get_vm_result_oop(oop_result);
 380   }
 381 
 382   bind(not_preempted);
 383 }
 384 
 385 static void pass_arg1(MacroAssembler* masm, Register arg) {
 386   if (c_rarg1 != arg ) {
 387     masm->mov(c_rarg1, arg);
 388   }
 389 }
 390 
 391 static void pass_arg2(MacroAssembler* masm, Register arg) {
 392   if (c_rarg2 != arg ) {
 393     masm->mov(c_rarg2, arg);
 394   }
 395 }
 396 
 397 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
 398                                          address entry_point,
 399                                          Register arg_1,
 400                                          bool check_exceptions) {
 401   pass_arg1(this, arg_1);
 402   call_VM_preemptable_helper(oop_result, entry_point, 1, check_exceptions);
 403 }
 404 
 405 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
 406                                          address entry_point,
 407                                          Register arg_1,
 408                                          Register arg_2,
 409                                          bool check_exceptions) {
 410   LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
 411   pass_arg2(this, arg_2);
 412   pass_arg1(this, arg_1);
 413   call_VM_preemptable_helper(oop_result, entry_point, 2, check_exceptions);
 414 }
 415 
 416 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
 417   lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
 418   call(rscratch1);
 419   if (is_native) {
 420     // On resume we need to set up stack as expected.
 421     push(dtos);
 422     push(ltos);
 423   }
 424 }
 425 
 426 void InterpreterMacroAssembler::check_and_handle_popframe() {
 427   if (JvmtiExport::can_pop_frame()) {
 428     Label L;
 429     // Initiate popframe handling only if it is not already being
 430     // processed.  If the flag has the popframe_processing bit set, it
 431     // means that this code is called *during* popframe handling - we
 432     // don't want to reenter.
 433     // This method is only called just after the call into the vm in
 434     // call_VM_base, so the arg registers are available.
 435     Register pop_cond = c_rarg0;
 436     movl(pop_cond, Address(r15_thread, JavaThread::popframe_condition_offset()));
 437     testl(pop_cond, JavaThread::popframe_pending_bit);
 438     jcc(Assembler::zero, L);
 439     testl(pop_cond, JavaThread::popframe_processing_bit);
 440     jcc(Assembler::notZero, L);
 441     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 442     // address of the same-named entrypoint in the generated interpreter code.
 443     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 444     jmp(rax);
 445     bind(L);
 446   }
 447 }
 448 
 449 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 450   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 451   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 452   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 453   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 454 
 455   switch (state) {
 456     case atos: movptr(rax, oop_addr);
 457                movptr(oop_addr, NULL_WORD);
 458                interp_verify_oop(rax, state);         break;
 459     case ltos: movptr(rax, val_addr);                 break;
 460     case btos:                                   // fall through
 461     case ztos:                                   // fall through
 462     case ctos:                                   // fall through
 463     case stos:                                   // fall through
 464     case itos: movl(rax, val_addr);                 break;
 465     case ftos: movflt(xmm0, val_addr);              break;
 466     case dtos: movdbl(xmm0, val_addr);              break;
 467     case vtos: /* nothing to do */                  break;
 468     default  : ShouldNotReachHere();
 469   }
 470 
 471   // Clean up tos value in the thread object
 472   movl(tos_addr, ilgl);
 473   movptr(val_addr, NULL_WORD);
 474 }
 475 
 476 
 477 void InterpreterMacroAssembler::check_and_handle_earlyret() {
 478   if (JvmtiExport::can_force_early_return()) {
 479     Label L;
 480     Register tmp = c_rarg0;
 481     Register rthread = r15_thread;
 482 
 483     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 484     testptr(tmp, tmp);
 485     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
 486 
 487     // Initiate earlyret handling only if it is not already being processed.
 488     // If the flag has the earlyret_processing bit set, it means that this code
 489     // is called *during* earlyret handling - we don't want to reenter.
 490     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
 491     cmpl(tmp, JvmtiThreadState::earlyret_pending);
 492     jcc(Assembler::notEqual, L);
 493 
 494     // Call Interpreter::remove_activation_early_entry() to get the address of the
 495     // same-named entrypoint in the generated interpreter code.
 496     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 497     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 498     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
 499     jmp(rax);
 500     bind(L);
 501   }
 502 }
 503 
 504 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
 505   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 506   load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
 507   bswapl(reg);
 508   shrl(reg, 16);
 509 }
 510 
 511 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 512                                                        int bcp_offset,
 513                                                        size_t index_size) {
 514   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 515   if (index_size == sizeof(u2)) {
 516     load_unsigned_short(index, Address(_bcp_register, bcp_offset));
 517   } else if (index_size == sizeof(u4)) {
 518     movl(index, Address(_bcp_register, bcp_offset));
 519   } else if (index_size == sizeof(u1)) {
 520     load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
 521   } else {
 522     ShouldNotReachHere();
 523   }
 524 }
 525 
 526 // Load object from cpool->resolved_references(index)
 527 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
 528                                                                  Register index,
 529                                                                  Register tmp) {
 530   assert_different_registers(result, index);
 531 
 532   get_constant_pool(result);
 533   // load pointer for resolved_references[] objArray
 534   movptr(result, Address(result, ConstantPool::cache_offset()));
 535   movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
 536   resolve_oop_handle(result, tmp);
 537   load_heap_oop(result, Address(result, index,
 538                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 539                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
 540 }
 541 
 542 // load cpool->resolved_klass_at(index)
 543 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
 544                                                              Register cpool,
 545                                                              Register index) {
 546   assert_different_registers(cpool, index);
 547 
 548   movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
 549   Register resolved_klasses = cpool;
 550   movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
 551   movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
 552 }
 553 
 554 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 555 // subtype of super_klass.
 556 //
 557 // Args:
 558 //      rax: superklass
 559 //      Rsub_klass: subklass
 560 //
 561 // Kills:
 562 //      rcx, rdi
 563 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 564                                                   Label& ok_is_subtype) {
 565   assert(Rsub_klass != rax, "rax holds superklass");
 566   LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
 567   LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
 568   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 569   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 570 
 571   // Profile the not-null value's klass.
 572   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 573 
 574   // Do the check.
 575   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 576 }
 577 
 578 
 579 // Java Expression Stack
 580 
 581 void InterpreterMacroAssembler::pop_ptr(Register r) {
 582   pop(r);
 583 }
 584 
 585 void InterpreterMacroAssembler::push_ptr(Register r) {
 586   push(r);
 587 }
 588 
 589 void InterpreterMacroAssembler::push_i(Register r) {
 590   push(r);
 591 }
 592 
 593 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
 594   push(r);
 595 }
 596 
 597 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 598   subptr(rsp, wordSize);
 599   movflt(Address(rsp, 0), r);
 600 }
 601 
 602 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 603   movflt(r, Address(rsp, 0));
 604   addptr(rsp, wordSize);
 605 }
 606 
 607 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 608   subptr(rsp, 2 * wordSize);
 609   movdbl(Address(rsp, 0), r);
 610 }
 611 
 612 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 613   movdbl(r, Address(rsp, 0));
 614   addptr(rsp, 2 * Interpreter::stackElementSize);
 615 }
 616 
 617 void InterpreterMacroAssembler::pop_i(Register r) {
 618   // XXX can't use pop currently, upper half non clean
 619   movl(r, Address(rsp, 0));
 620   addptr(rsp, wordSize);
 621 }
 622 
 623 void InterpreterMacroAssembler::pop_l(Register r) {
 624   movq(r, Address(rsp, 0));
 625   addptr(rsp, 2 * Interpreter::stackElementSize);
 626 }
 627 
 628 void InterpreterMacroAssembler::push_l(Register r) {
 629   subptr(rsp, 2 * wordSize);
 630   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r         );
 631   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
 632 }
 633 
 634 void InterpreterMacroAssembler::pop(TosState state) {
 635   switch (state) {
 636   case atos: pop_ptr();                 break;
 637   case btos:
 638   case ztos:
 639   case ctos:
 640   case stos:
 641   case itos: pop_i();                   break;
 642   case ltos: pop_l();                   break;
 643   case ftos: pop_f(xmm0);               break;
 644   case dtos: pop_d(xmm0);               break;
 645   case vtos: /* nothing to do */        break;
 646   default:   ShouldNotReachHere();
 647   }
 648   interp_verify_oop(rax, state);
 649 }
 650 
 651 void InterpreterMacroAssembler::push(TosState state) {
 652   interp_verify_oop(rax, state);
 653   switch (state) {
 654   case atos: push_ptr();                break;
 655   case btos:
 656   case ztos:
 657   case ctos:
 658   case stos:
 659   case itos: push_i();                  break;
 660   case ltos: push_l();                  break;
 661   case ftos: push_f(xmm0);              break;
 662   case dtos: push_d(xmm0);              break;
 663   case vtos: /* nothing to do */        break;
 664   default  : ShouldNotReachHere();
 665   }
 666 }
 667 
 668 // Helpers for swap and dup
 669 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 670   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 671 }
 672 
 673 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 674   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 675 }
 676 
 677 
 678 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 679   // set sender sp
 680   lea(_bcp_register, Address(rsp, wordSize));
 681   // record last_sp
 682   mov(rcx, _bcp_register);
 683   subptr(rcx, rbp);
 684   sarptr(rcx, LogBytesPerWord);
 685   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
 686 }
 687 
 688 
 689 // Jump to from_interpreted entry of a call unless single stepping is possible
 690 // in this thread in which case we must call the i2i entry
 691 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 692   prepare_to_jump_from_interpreted();
 693 
 694   if (JvmtiExport::can_post_interpreter_events()) {
 695     Label run_compiled_code;
 696     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 697     // compiled code in threads for which the event is enabled.  Check here for
 698     // interp_only_mode if these events CAN be enabled.
 699     // interp_only is an int, on little endian it is sufficient to test the byte only
 700     // Is a cmpl faster?
 701     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
 702     jccb(Assembler::zero, run_compiled_code);
 703     jmp(Address(method, Method::interpreter_entry_offset()));
 704     bind(run_compiled_code);
 705   }
 706 
 707   jmp(Address(method, Method::from_interpreted_offset()));
 708 }
 709 
 710 // The following two routines provide a hook so that an implementation
 711 // can schedule the dispatch in two parts.  x86 does not do this.
 712 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 713   // Nothing x86 specific to be done here
 714 }
 715 
 716 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 717   dispatch_next(state, step);
 718 }
 719 
 720 void InterpreterMacroAssembler::dispatch_base(TosState state,
 721                                               address* table,
 722                                               bool verifyoop,
 723                                               bool generate_poll) {
 724   if (VerifyActivationFrameSize) {
 725     Label L;
 726     mov(rcx, rbp);
 727     subptr(rcx, rsp);
 728     int32_t min_frame_size =
 729       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 730       wordSize;
 731     cmpptr(rcx, min_frame_size);
 732     jcc(Assembler::greaterEqual, L);
 733     stop("broken stack frame");
 734     bind(L);
 735   }
 736   if (verifyoop) {
 737     interp_verify_oop(rax, state);
 738   }
 739 
 740   address* const safepoint_table = Interpreter::safept_table(state);
 741   Label no_safepoint, dispatch;
 742   if (table != safepoint_table && generate_poll) {
 743     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 744     testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 745 
 746     jccb(Assembler::zero, no_safepoint);
 747     lea(rscratch1, ExternalAddress((address)safepoint_table));
 748     jmpb(dispatch);
 749   }
 750 
 751   bind(no_safepoint);
 752   lea(rscratch1, ExternalAddress((address)table));
 753   bind(dispatch);
 754   jmp(Address(rscratch1, rbx, Address::times_8));
 755 }
 756 
 757 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 758   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 759 }
 760 
 761 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 762   dispatch_base(state, Interpreter::normal_table(state));
 763 }
 764 
 765 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 766   dispatch_base(state, Interpreter::normal_table(state), false);
 767 }
 768 
 769 
 770 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 771   // load next bytecode (load before advancing _bcp_register to prevent AGI)
 772   load_unsigned_byte(rbx, Address(_bcp_register, step));
 773   // advance _bcp_register
 774   increment(_bcp_register, step);
 775   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 776 }
 777 
 778 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 779   // load current bytecode
 780   load_unsigned_byte(rbx, Address(_bcp_register, 0));
 781   dispatch_base(state, table);
 782 }
 783 
 784 void InterpreterMacroAssembler::narrow(Register result) {
 785 
 786   // Get method->_constMethod->_result_type
 787   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 788   movptr(rcx, Address(rcx, Method::const_offset()));
 789   load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
 790 
 791   Label done, notBool, notByte, notChar;
 792 
 793   // common case first
 794   cmpl(rcx, T_INT);
 795   jcc(Assembler::equal, done);
 796 
 797   // mask integer result to narrower return type.
 798   cmpl(rcx, T_BOOLEAN);
 799   jcc(Assembler::notEqual, notBool);
 800   andl(result, 0x1);
 801   jmp(done);
 802 
 803   bind(notBool);
 804   cmpl(rcx, T_BYTE);
 805   jcc(Assembler::notEqual, notByte);
 806   movsbl(result, result);
 807   jmp(done);
 808 
 809   bind(notByte);
 810   cmpl(rcx, T_CHAR);
 811   jcc(Assembler::notEqual, notChar);
 812   movzwl(result, result);
 813   jmp(done);
 814 
 815   bind(notChar);
 816   // cmpl(rcx, T_SHORT);  // all that's left
 817   // jcc(Assembler::notEqual, done);
 818   movswl(result, result);
 819 
 820   // Nothing to do for T_INT
 821   bind(done);
 822 }
 823 
 824 // remove activation
 825 //
 826 // Apply stack watermark barrier.
 827 // Unlock the receiver if this is a synchronized method.
 828 // Unlock any Java monitors from synchronized blocks.
 829 // Remove the activation from the stack.
 830 //
 831 // If there are locked Java monitors
 832 //    If throw_monitor_exception
 833 //       throws IllegalMonitorStateException
 834 //    Else if install_monitor_exception
 835 //       installs IllegalMonitorStateException
 836 //    Else
 837 //       no error processing
 838 void InterpreterMacroAssembler::remove_activation(
 839         TosState state,
 840         Register ret_addr,
 841         bool throw_monitor_exception,
 842         bool install_monitor_exception,
 843         bool notify_jvmdi) {
 844   // Note: Registers rdx xmm0 may be in use for the
 845   // result check if synchronized method
 846   Label unlocked, unlock, no_unlock;
 847 
 848   const Register rthread = r15_thread;
 849   const Register robj    = c_rarg1;
 850   const Register rmon    = c_rarg1;
 851 
 852   // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
 853   // that would normally not be safe to use. Such bad returns into unsafe territory of
 854   // the stack, will call InterpreterRuntime::at_unwind.
 855   Label slow_path;
 856   Label fast_path;
 857   safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
 858   jmp(fast_path);
 859   bind(slow_path);
 860   push(state);
 861   set_last_Java_frame(noreg, rbp, (address)pc(), rscratch1);
 862   super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
 863   reset_last_Java_frame(true);
 864   pop(state);
 865   bind(fast_path);
 866 
 867   // get the value of _do_not_unlock_if_synchronized into rdx
 868   const Address do_not_unlock_if_synchronized(rthread,
 869     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 870   movbool(rbx, do_not_unlock_if_synchronized);
 871   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 872 
 873  // get method access flags
 874   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 875   load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset()));
 876   testl(rcx, JVM_ACC_SYNCHRONIZED);
 877   jcc(Assembler::zero, unlocked);
 878 
 879   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 880   // is set.
 881   testbool(rbx);
 882   jcc(Assembler::notZero, no_unlock);
 883 
 884   // unlock monitor
 885   push(state); // save result
 886 
 887   // BasicObjectLock will be first in list, since this is a
 888   // synchronized method. However, need to check that the object has
 889   // not been unlocked by an explicit monitorexit bytecode.
 890   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 891                         wordSize - (int) sizeof(BasicObjectLock));
 892   // We use c_rarg1/rdx so that if we go slow path it will be the correct
 893   // register for unlock_object to pass to VM directly
 894   lea(robj, monitor); // address of first monitor
 895 
 896   movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
 897   testptr(rax, rax);
 898   jcc(Assembler::notZero, unlock);
 899 
 900   pop(state);
 901   if (throw_monitor_exception) {
 902     // Entry already unlocked, need to throw exception
 903     call_VM(noreg, CAST_FROM_FN_PTR(address,
 904                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 905     should_not_reach_here();
 906   } else {
 907     // Monitor already unlocked during a stack unroll. If requested,
 908     // install an illegal_monitor_state_exception.  Continue with
 909     // stack unrolling.
 910     if (install_monitor_exception) {
 911       call_VM(noreg, CAST_FROM_FN_PTR(address,
 912                      InterpreterRuntime::new_illegal_monitor_state_exception));
 913     }
 914     jmp(unlocked);
 915   }
 916 
 917   bind(unlock);
 918   unlock_object(robj);
 919   pop(state);
 920 
 921   // Check that for block-structured locking (i.e., that all locked
 922   // objects has been unlocked)
 923   bind(unlocked);
 924 
 925   // rax, rdx: Might contain return value
 926 
 927   // Check that all monitors are unlocked
 928   {
 929     Label loop, exception, entry, restart;
 930     const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 931     const Address monitor_block_top(
 932         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 933     const Address monitor_block_bot(
 934         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
 935 
 936     bind(restart);
 937     // We use c_rarg1 so that if we go slow path it will be the correct
 938     // register for unlock_object to pass to VM directly
 939     movptr(rmon, monitor_block_top); // derelativize pointer
 940     lea(rmon, Address(rbp, rmon, Address::times_ptr));
 941     // c_rarg1 points to current entry, starting with top-most entry
 942 
 943     lea(rbx, monitor_block_bot);  // points to word before bottom of
 944                                   // monitor block
 945     jmp(entry);
 946 
 947     // Entry already locked, need to throw exception
 948     bind(exception);
 949 
 950     if (throw_monitor_exception) {
 951       // Throw exception
 952       MacroAssembler::call_VM(noreg,
 953                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 954                                    throw_illegal_monitor_state_exception));
 955       should_not_reach_here();
 956     } else {
 957       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 958       // Unlock does not block, so don't have to worry about the frame.
 959       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 960 
 961       push(state);
 962       mov(robj, rmon);   // nop if robj and rmon are the same
 963       unlock_object(robj);
 964       pop(state);
 965 
 966       if (install_monitor_exception) {
 967         call_VM(noreg, CAST_FROM_FN_PTR(address,
 968                                         InterpreterRuntime::
 969                                         new_illegal_monitor_state_exception));
 970       }
 971 
 972       jmp(restart);
 973     }
 974 
 975     bind(loop);
 976     // check if current entry is used
 977     cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
 978     jcc(Assembler::notEqual, exception);
 979 
 980     addptr(rmon, entry_size); // otherwise advance to next entry
 981     bind(entry);
 982     cmpptr(rmon, rbx); // check if bottom reached
 983     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
 984   }
 985 
 986   bind(no_unlock);
 987 
 988   // jvmti support
 989   if (notify_jvmdi) {
 990     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 991   } else {
 992     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 993   }
 994 
 995   // remove activation
 996   // get sender sp
 997   movptr(rbx,
 998          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
 999   if (StackReservedPages > 0) {
1000     // testing if reserved zone needs to be re-enabled
1001     Register rthread = r15_thread;
1002     Label no_reserved_zone_enabling;
1003 
1004     // check if already enabled - if so no re-enabling needed
1005     assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1006     cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1007     jcc(Assembler::equal, no_reserved_zone_enabling);
1008 
1009     cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1010     jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1011 
1012     call_VM_leaf(
1013       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1014     call_VM(noreg, CAST_FROM_FN_PTR(address,
1015                    InterpreterRuntime::throw_delayed_StackOverflowError));
1016     should_not_reach_here();
1017 
1018     bind(no_reserved_zone_enabling);
1019   }
1020   leave();                           // remove frame anchor
1021   pop(ret_addr);                     // get return address
1022   mov(rsp, rbx);                     // set sp to sender sp
1023   pop_cont_fastpath();
1024 }
1025 
1026 void InterpreterMacroAssembler::get_method_counters(Register method,
1027                                                     Register mcs, Label& skip) {
1028   Label has_counters;
1029   movptr(mcs, Address(method, Method::method_counters_offset()));
1030   testptr(mcs, mcs);
1031   jcc(Assembler::notZero, has_counters);
1032   call_VM(noreg, CAST_FROM_FN_PTR(address,
1033           InterpreterRuntime::build_method_counters), method);
1034   movptr(mcs, Address(method,Method::method_counters_offset()));
1035   testptr(mcs, mcs);
1036   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1037   bind(has_counters);
1038 }
1039 
1040 
1041 // Lock object
1042 //
1043 // Args:
1044 //      rdx, c_rarg1: BasicObjectLock to be used for locking
1045 //
1046 // Kills:
1047 //      rax, rbx
1048 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1049   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1050 
1051   if (LockingMode == LM_MONITOR) {
1052     call_VM_preemptable(noreg,
1053             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1054             lock_reg);
1055   } else {
1056     Label count_locking, done, slow_case;
1057 
1058     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1059     const Register tmp_reg = rbx;
1060     const Register obj_reg = c_rarg3; // Will contain the oop
1061     const Register rklass_decode_tmp = rscratch1;
1062 
1063     const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1064     const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1065     const int mark_offset = lock_offset +
1066                             BasicLock::displaced_header_offset_in_bytes();
1067 
1068     // Load object pointer into obj_reg
1069     movptr(obj_reg, Address(lock_reg, obj_offset));
1070 
1071     if (DiagnoseSyncOnValueBasedClasses != 0) {
1072       load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1073       testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1074       jcc(Assembler::notZero, slow_case);
1075     }
1076 
1077     if (LockingMode == LM_LIGHTWEIGHT) {
1078       lightweight_lock(lock_reg, obj_reg, swap_reg, tmp_reg, slow_case);
1079     } else if (LockingMode == LM_LEGACY) {
1080       // Load immediate 1 into swap_reg %rax
1081       movl(swap_reg, 1);
1082 
1083       // Load (object->mark() | 1) into swap_reg %rax
1084       orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1085 
1086       // Save (object->mark() | 1) into BasicLock's displaced header
1087       movptr(Address(lock_reg, mark_offset), swap_reg);
1088 
1089       assert(lock_offset == 0,
1090              "displaced header must be first word in BasicObjectLock");
1091 
1092       lock();
1093       cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1094       jcc(Assembler::zero, count_locking);
1095 
1096       const int zero_bits = 7;
1097 
1098       // Fast check for recursive lock.
1099       //
1100       // Can apply the optimization only if this is a stack lock
1101       // allocated in this thread. For efficiency, we can focus on
1102       // recently allocated stack locks (instead of reading the stack
1103       // base and checking whether 'mark' points inside the current
1104       // thread stack):
1105       //  1) (mark & zero_bits) == 0, and
1106       //  2) rsp <= mark < mark + os::pagesize()
1107       //
1108       // Warning: rsp + os::pagesize can overflow the stack base. We must
1109       // neither apply the optimization for an inflated lock allocated
1110       // just above the thread stack (this is why condition 1 matters)
1111       // nor apply the optimization if the stack lock is inside the stack
1112       // of another thread. The latter is avoided even in case of overflow
1113       // because we have guard pages at the end of all stacks. Hence, if
1114       // we go over the stack base and hit the stack of another thread,
1115       // this should not be in a writeable area that could contain a
1116       // stack lock allocated by that thread. As a consequence, a stack
1117       // lock less than page size away from rsp is guaranteed to be
1118       // owned by the current thread.
1119       //
1120       // These 3 tests can be done by evaluating the following
1121       // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1122       // assuming both stack pointer and pagesize have their
1123       // least significant bits clear.
1124       // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1125       subptr(swap_reg, rsp);
1126       andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1127 
1128       // Save the test result, for recursive case, the result is zero
1129       movptr(Address(lock_reg, mark_offset), swap_reg);
1130       jcc(Assembler::notZero, slow_case);
1131 
1132       bind(count_locking);
1133       inc_held_monitor_count();
1134     }
1135     jmp(done);
1136 
1137     bind(slow_case);
1138 
1139     // Call the runtime routine for slow case
1140     call_VM_preemptable(noreg,
1141             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1142             lock_reg);
1143     bind(done);
1144   }
1145 }
1146 
1147 
1148 // Unlocks an object. Used in monitorexit bytecode and
1149 // remove_activation.  Throws an IllegalMonitorException if object is
1150 // not locked by current thread.
1151 //
1152 // Args:
1153 //      rdx, c_rarg1: BasicObjectLock for lock
1154 //
1155 // Kills:
1156 //      rax
1157 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1158 //      rscratch1 (scratch reg)
1159 // rax, rbx, rcx, rdx
1160 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1161   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1162 
1163   if (LockingMode == LM_MONITOR) {
1164     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1165   } else {
1166     Label count_locking, done, slow_case;
1167 
1168     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
1169     const Register header_reg = c_rarg2;  // Will contain the old oopMark
1170     const Register obj_reg    = c_rarg3;  // Will contain the oop
1171 
1172     save_bcp(); // Save in case of exception
1173 
1174     if (LockingMode != LM_LIGHTWEIGHT) {
1175       // Convert from BasicObjectLock structure to object and BasicLock
1176       // structure Store the BasicLock address into %rax
1177       lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1178     }
1179 
1180     // Load oop into obj_reg(%c_rarg3)
1181     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1182 
1183     // Free entry
1184     movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1185 
1186     if (LockingMode == LM_LIGHTWEIGHT) {
1187       lightweight_unlock(obj_reg, swap_reg, header_reg, slow_case);
1188     } else if (LockingMode == LM_LEGACY) {
1189       // Load the old header from BasicLock structure
1190       movptr(header_reg, Address(swap_reg,
1191                                  BasicLock::displaced_header_offset_in_bytes()));
1192 
1193       // Test for recursion
1194       testptr(header_reg, header_reg);
1195 
1196       // zero for recursive case
1197       jcc(Assembler::zero, count_locking);
1198 
1199       // Atomic swap back the old header
1200       lock();
1201       cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1202 
1203       // zero for simple unlock of a stack-lock case
1204       jcc(Assembler::notZero, slow_case);
1205 
1206       bind(count_locking);
1207       dec_held_monitor_count();
1208     }
1209     jmp(done);
1210 
1211     bind(slow_case);
1212     // Call the runtime routine for slow case.
1213     movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1214     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1215 
1216     bind(done);
1217 
1218     restore_bcp();
1219   }
1220 }
1221 
1222 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1223                                                          Label& zero_continue) {
1224   assert(ProfileInterpreter, "must be profiling interpreter");
1225   movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1226   testptr(mdp, mdp);
1227   jcc(Assembler::zero, zero_continue);
1228 }
1229 
1230 
1231 // Set the method data pointer for the current bcp.
1232 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1233   assert(ProfileInterpreter, "must be profiling interpreter");
1234   Label set_mdp;
1235   push(rax);
1236   push(rbx);
1237 
1238   get_method(rbx);
1239   // Test MDO to avoid the call if it is null.
1240   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1241   testptr(rax, rax);
1242   jcc(Assembler::zero, set_mdp);
1243   // rbx: method
1244   // _bcp_register: bcp
1245   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1246   // rax: mdi
1247   // mdo is guaranteed to be non-zero here, we checked for it before the call.
1248   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1249   addptr(rbx, in_bytes(MethodData::data_offset()));
1250   addptr(rax, rbx);
1251   bind(set_mdp);
1252   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1253   pop(rbx);
1254   pop(rax);
1255 }
1256 
1257 void InterpreterMacroAssembler::verify_method_data_pointer() {
1258   assert(ProfileInterpreter, "must be profiling interpreter");
1259 #ifdef ASSERT
1260   Label verify_continue;
1261   push(rax);
1262   push(rbx);
1263   Register arg3_reg = c_rarg3;
1264   Register arg2_reg = c_rarg2;
1265   push(arg3_reg);
1266   push(arg2_reg);
1267   test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1268   get_method(rbx);
1269 
1270   // If the mdp is valid, it will point to a DataLayout header which is
1271   // consistent with the bcp.  The converse is highly probable also.
1272   load_unsigned_short(arg2_reg,
1273                       Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1274   addptr(arg2_reg, Address(rbx, Method::const_offset()));
1275   lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1276   cmpptr(arg2_reg, _bcp_register);
1277   jcc(Assembler::equal, verify_continue);
1278   // rbx: method
1279   // _bcp_register: bcp
1280   // c_rarg3: mdp
1281   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1282                rbx, _bcp_register, arg3_reg);
1283   bind(verify_continue);
1284   pop(arg2_reg);
1285   pop(arg3_reg);
1286   pop(rbx);
1287   pop(rax);
1288 #endif // ASSERT
1289 }
1290 
1291 
1292 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1293                                                 int constant,
1294                                                 Register value) {
1295   assert(ProfileInterpreter, "must be profiling interpreter");
1296   Address data(mdp_in, constant);
1297   movptr(data, value);
1298 }
1299 
1300 
1301 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1302                                                       int constant,
1303                                                       bool decrement) {
1304   // Counter address
1305   Address data(mdp_in, constant);
1306 
1307   increment_mdp_data_at(data, decrement);
1308 }
1309 
1310 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1311                                                       bool decrement) {
1312   assert(ProfileInterpreter, "must be profiling interpreter");
1313   // %%% this does 64bit counters at best it is wasting space
1314   // at worst it is a rare bug when counters overflow
1315 
1316   if (decrement) {
1317     // Decrement the register.  Set condition codes.
1318     addptr(data, -DataLayout::counter_increment);
1319     // If the decrement causes the counter to overflow, stay negative
1320     Label L;
1321     jcc(Assembler::negative, L);
1322     addptr(data, DataLayout::counter_increment);
1323     bind(L);
1324   } else {
1325     assert(DataLayout::counter_increment == 1,
1326            "flow-free idiom only works with 1");
1327     // Increment the register.  Set carry flag.
1328     addptr(data, DataLayout::counter_increment);
1329     // If the increment causes the counter to overflow, pull back by 1.
1330     sbbptr(data, 0);
1331   }
1332 }
1333 
1334 
1335 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1336                                                       Register reg,
1337                                                       int constant,
1338                                                       bool decrement) {
1339   Address data(mdp_in, reg, Address::times_1, constant);
1340 
1341   increment_mdp_data_at(data, decrement);
1342 }
1343 
1344 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1345                                                 int flag_byte_constant) {
1346   assert(ProfileInterpreter, "must be profiling interpreter");
1347   int header_offset = in_bytes(DataLayout::flags_offset());
1348   int header_bits = flag_byte_constant;
1349   // Set the flag
1350   orb(Address(mdp_in, header_offset), header_bits);
1351 }
1352 
1353 
1354 
1355 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1356                                                  int offset,
1357                                                  Register value,
1358                                                  Register test_value_out,
1359                                                  Label& not_equal_continue) {
1360   assert(ProfileInterpreter, "must be profiling interpreter");
1361   if (test_value_out == noreg) {
1362     cmpptr(value, Address(mdp_in, offset));
1363   } else {
1364     // Put the test value into a register, so caller can use it:
1365     movptr(test_value_out, Address(mdp_in, offset));
1366     cmpptr(test_value_out, value);
1367   }
1368   jcc(Assembler::notEqual, not_equal_continue);
1369 }
1370 
1371 
1372 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1373                                                      int offset_of_disp) {
1374   assert(ProfileInterpreter, "must be profiling interpreter");
1375   Address disp_address(mdp_in, offset_of_disp);
1376   addptr(mdp_in, disp_address);
1377   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1378 }
1379 
1380 
1381 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1382                                                      Register reg,
1383                                                      int offset_of_disp) {
1384   assert(ProfileInterpreter, "must be profiling interpreter");
1385   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1386   addptr(mdp_in, disp_address);
1387   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1388 }
1389 
1390 
1391 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1392                                                        int constant) {
1393   assert(ProfileInterpreter, "must be profiling interpreter");
1394   addptr(mdp_in, constant);
1395   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1396 }
1397 
1398 
1399 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1400   assert(ProfileInterpreter, "must be profiling interpreter");
1401   push(return_bci); // save/restore across call_VM
1402   call_VM(noreg,
1403           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1404           return_bci);
1405   pop(return_bci);
1406 }
1407 
1408 
1409 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1410                                                      Register bumped_count) {
1411   if (ProfileInterpreter) {
1412     Label profile_continue;
1413 
1414     // If no method data exists, go to profile_continue.
1415     // Otherwise, assign to mdp
1416     test_method_data_pointer(mdp, profile_continue);
1417 
1418     // We are taking a branch.  Increment the taken count.
1419     // We inline increment_mdp_data_at to return bumped_count in a register
1420     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1421     Address data(mdp, in_bytes(JumpData::taken_offset()));
1422     movptr(bumped_count, data);
1423     assert(DataLayout::counter_increment == 1,
1424             "flow-free idiom only works with 1");
1425     addptr(bumped_count, DataLayout::counter_increment);
1426     sbbptr(bumped_count, 0);
1427     movptr(data, bumped_count); // Store back out
1428 
1429     // The method data pointer needs to be updated to reflect the new target.
1430     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1431     bind(profile_continue);
1432   }
1433 }
1434 
1435 
1436 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1437   if (ProfileInterpreter) {
1438     Label profile_continue;
1439 
1440     // If no method data exists, go to profile_continue.
1441     test_method_data_pointer(mdp, profile_continue);
1442 
1443     // We are taking a branch.  Increment the not taken count.
1444     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1445 
1446     // The method data pointer needs to be updated to correspond to
1447     // the next bytecode
1448     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1449     bind(profile_continue);
1450   }
1451 }
1452 
1453 void InterpreterMacroAssembler::profile_call(Register mdp) {
1454   if (ProfileInterpreter) {
1455     Label profile_continue;
1456 
1457     // If no method data exists, go to profile_continue.
1458     test_method_data_pointer(mdp, profile_continue);
1459 
1460     // We are making a call.  Increment the count.
1461     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1462 
1463     // The method data pointer needs to be updated to reflect the new target.
1464     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1465     bind(profile_continue);
1466   }
1467 }
1468 
1469 
1470 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1471   if (ProfileInterpreter) {
1472     Label profile_continue;
1473 
1474     // If no method data exists, go to profile_continue.
1475     test_method_data_pointer(mdp, profile_continue);
1476 
1477     // We are making a call.  Increment the count.
1478     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1479 
1480     // The method data pointer needs to be updated to reflect the new target.
1481     update_mdp_by_constant(mdp,
1482                            in_bytes(VirtualCallData::
1483                                     virtual_call_data_size()));
1484     bind(profile_continue);
1485   }
1486 }
1487 
1488 
1489 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1490                                                      Register mdp,
1491                                                      Register reg2,
1492                                                      bool receiver_can_be_null) {
1493   if (ProfileInterpreter) {
1494     Label profile_continue;
1495 
1496     // If no method data exists, go to profile_continue.
1497     test_method_data_pointer(mdp, profile_continue);
1498 
1499     Label skip_receiver_profile;
1500     if (receiver_can_be_null) {
1501       Label not_null;
1502       testptr(receiver, receiver);
1503       jccb(Assembler::notZero, not_null);
1504       // We are making a call.  Increment the count for null receiver.
1505       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1506       jmp(skip_receiver_profile);
1507       bind(not_null);
1508     }
1509 
1510     // Record the receiver type.
1511     record_klass_in_profile(receiver, mdp, reg2, true);
1512     bind(skip_receiver_profile);
1513 
1514     // The method data pointer needs to be updated to reflect the new target.
1515     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1516     bind(profile_continue);
1517   }
1518 }
1519 
1520 // This routine creates a state machine for updating the multi-row
1521 // type profile at a virtual call site (or other type-sensitive bytecode).
1522 // The machine visits each row (of receiver/count) until the receiver type
1523 // is found, or until it runs out of rows.  At the same time, it remembers
1524 // the location of the first empty row.  (An empty row records null for its
1525 // receiver, and can be allocated for a newly-observed receiver type.)
1526 // Because there are two degrees of freedom in the state, a simple linear
1527 // search will not work; it must be a decision tree.  Hence this helper
1528 // function is recursive, to generate the required tree structured code.
1529 // It's the interpreter, so we are trading off code space for speed.
1530 // See below for example code.
1531 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1532                                         Register receiver, Register mdp,
1533                                         Register reg2, int start_row,
1534                                         Label& done, bool is_virtual_call) {
1535   if (TypeProfileWidth == 0) {
1536     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1537   } else {
1538     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1539                                   &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1540   }
1541 }
1542 
1543 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1544                                                               Label& done, int total_rows,
1545                                                               OffsetFunction item_offset_fn,
1546                                                               OffsetFunction item_count_offset_fn) {
1547   int last_row = total_rows - 1;
1548   assert(start_row <= last_row, "must be work left to do");
1549   // Test this row for both the item and for null.
1550   // Take any of three different outcomes:
1551   //   1. found item => increment count and goto done
1552   //   2. found null => keep looking for case 1, maybe allocate this cell
1553   //   3. found something else => keep looking for cases 1 and 2
1554   // Case 3 is handled by a recursive call.
1555   for (int row = start_row; row <= last_row; row++) {
1556     Label next_test;
1557     bool test_for_null_also = (row == start_row);
1558 
1559     // See if the item is item[n].
1560     int item_offset = in_bytes(item_offset_fn(row));
1561     test_mdp_data_at(mdp, item_offset, item,
1562                      (test_for_null_also ? reg2 : noreg),
1563                      next_test);
1564     // (Reg2 now contains the item from the CallData.)
1565 
1566     // The item is item[n].  Increment count[n].
1567     int count_offset = in_bytes(item_count_offset_fn(row));
1568     increment_mdp_data_at(mdp, count_offset);
1569     jmp(done);
1570     bind(next_test);
1571 
1572     if (test_for_null_also) {
1573       // Failed the equality check on item[n]...  Test for null.
1574       testptr(reg2, reg2);
1575       if (start_row == last_row) {
1576         // The only thing left to do is handle the null case.
1577         Label found_null;
1578         jccb(Assembler::zero, found_null);
1579         // Item did not match any saved item and there is no empty row for it.
1580         // Increment total counter to indicate polymorphic case.
1581         increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1582         jmp(done);
1583         bind(found_null);
1584         break;
1585       }
1586       Label found_null;
1587       // Since null is rare, make it be the branch-taken case.
1588       jcc(Assembler::zero, found_null);
1589 
1590       // Put all the "Case 3" tests here.
1591       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1592                                     item_offset_fn, item_count_offset_fn);
1593 
1594       // Found a null.  Keep searching for a matching item,
1595       // but remember that this is an empty (unused) slot.
1596       bind(found_null);
1597     }
1598   }
1599 
1600   // In the fall-through case, we found no matching item, but we
1601   // observed the item[start_row] is null.
1602 
1603   // Fill in the item field and increment the count.
1604   int item_offset = in_bytes(item_offset_fn(start_row));
1605   set_mdp_data_at(mdp, item_offset, item);
1606   int count_offset = in_bytes(item_count_offset_fn(start_row));
1607   movl(reg2, DataLayout::counter_increment);
1608   set_mdp_data_at(mdp, count_offset, reg2);
1609   if (start_row > 0) {
1610     jmp(done);
1611   }
1612 }
1613 
1614 // Example state machine code for three profile rows:
1615 //   // main copy of decision tree, rooted at row[1]
1616 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1617 //   if (row[0].rec != nullptr) {
1618 //     // inner copy of decision tree, rooted at row[1]
1619 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1620 //     if (row[1].rec != nullptr) {
1621 //       // degenerate decision tree, rooted at row[2]
1622 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1623 //       if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1624 //       row[2].init(rec); goto done;
1625 //     } else {
1626 //       // remember row[1] is empty
1627 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1628 //       row[1].init(rec); goto done;
1629 //     }
1630 //   } else {
1631 //     // remember row[0] is empty
1632 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1633 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1634 //     row[0].init(rec); goto done;
1635 //   }
1636 //   done:
1637 
1638 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1639                                                         Register mdp, Register reg2,
1640                                                         bool is_virtual_call) {
1641   assert(ProfileInterpreter, "must be profiling");
1642   Label done;
1643 
1644   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1645 
1646   bind (done);
1647 }
1648 
1649 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1650                                             Register mdp) {
1651   if (ProfileInterpreter) {
1652     Label profile_continue;
1653     uint row;
1654 
1655     // If no method data exists, go to profile_continue.
1656     test_method_data_pointer(mdp, profile_continue);
1657 
1658     // Update the total ret count.
1659     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1660 
1661     for (row = 0; row < RetData::row_limit(); row++) {
1662       Label next_test;
1663 
1664       // See if return_bci is equal to bci[n]:
1665       test_mdp_data_at(mdp,
1666                        in_bytes(RetData::bci_offset(row)),
1667                        return_bci, noreg,
1668                        next_test);
1669 
1670       // return_bci is equal to bci[n].  Increment the count.
1671       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1672 
1673       // The method data pointer needs to be updated to reflect the new target.
1674       update_mdp_by_offset(mdp,
1675                            in_bytes(RetData::bci_displacement_offset(row)));
1676       jmp(profile_continue);
1677       bind(next_test);
1678     }
1679 
1680     update_mdp_for_ret(return_bci);
1681 
1682     bind(profile_continue);
1683   }
1684 }
1685 
1686 
1687 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1688   if (ProfileInterpreter) {
1689     Label profile_continue;
1690 
1691     // If no method data exists, go to profile_continue.
1692     test_method_data_pointer(mdp, profile_continue);
1693 
1694     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1695 
1696     // The method data pointer needs to be updated.
1697     int mdp_delta = in_bytes(BitData::bit_data_size());
1698     if (TypeProfileCasts) {
1699       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1700     }
1701     update_mdp_by_constant(mdp, mdp_delta);
1702 
1703     bind(profile_continue);
1704   }
1705 }
1706 
1707 
1708 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1709   if (ProfileInterpreter) {
1710     Label profile_continue;
1711 
1712     // If no method data exists, go to profile_continue.
1713     test_method_data_pointer(mdp, profile_continue);
1714 
1715     // The method data pointer needs to be updated.
1716     int mdp_delta = in_bytes(BitData::bit_data_size());
1717     if (TypeProfileCasts) {
1718       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1719 
1720       // Record the object type.
1721       record_klass_in_profile(klass, mdp, reg2, false);
1722     }
1723     update_mdp_by_constant(mdp, mdp_delta);
1724 
1725     bind(profile_continue);
1726   }
1727 }
1728 
1729 
1730 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1731   if (ProfileInterpreter) {
1732     Label profile_continue;
1733 
1734     // If no method data exists, go to profile_continue.
1735     test_method_data_pointer(mdp, profile_continue);
1736 
1737     // Update the default case count
1738     increment_mdp_data_at(mdp,
1739                           in_bytes(MultiBranchData::default_count_offset()));
1740 
1741     // The method data pointer needs to be updated.
1742     update_mdp_by_offset(mdp,
1743                          in_bytes(MultiBranchData::
1744                                   default_displacement_offset()));
1745 
1746     bind(profile_continue);
1747   }
1748 }
1749 
1750 
1751 void InterpreterMacroAssembler::profile_switch_case(Register index,
1752                                                     Register mdp,
1753                                                     Register reg2) {
1754   if (ProfileInterpreter) {
1755     Label profile_continue;
1756 
1757     // If no method data exists, go to profile_continue.
1758     test_method_data_pointer(mdp, profile_continue);
1759 
1760     // Build the base (index * per_case_size_in_bytes()) +
1761     // case_array_offset_in_bytes()
1762     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1763     imulptr(index, reg2); // XXX l ?
1764     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1765 
1766     // Update the case count
1767     increment_mdp_data_at(mdp,
1768                           index,
1769                           in_bytes(MultiBranchData::relative_count_offset()));
1770 
1771     // The method data pointer needs to be updated.
1772     update_mdp_by_offset(mdp,
1773                          index,
1774                          in_bytes(MultiBranchData::
1775                                   relative_displacement_offset()));
1776 
1777     bind(profile_continue);
1778   }
1779 }
1780 
1781 
1782 
1783 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1784   if (state == atos) {
1785     MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1786   }
1787 }
1788 
1789 
1790 // Jump if ((*counter_addr += increment) & mask) == 0
1791 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1792                                                         Register scratch, Label* where) {
1793   // This update is actually not atomic and can lose a number of updates
1794   // under heavy contention, but the alternative of using the (contended)
1795   // atomic update here penalizes profiling paths too much.
1796   movl(scratch, counter_addr);
1797   incrementl(scratch, InvocationCounter::count_increment);
1798   movl(counter_addr, scratch);
1799   andl(scratch, mask);
1800   if (where != nullptr) {
1801     jcc(Assembler::zero, *where);
1802   }
1803 }
1804 
1805 void InterpreterMacroAssembler::notify_method_entry() {
1806   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1807   // track stack depth.  If it is possible to enter interp_only_mode we add
1808   // the code to check if the event should be sent.
1809   Register rthread = r15_thread;
1810   Register rarg = c_rarg1;
1811   if (JvmtiExport::can_post_interpreter_events()) {
1812     Label L;
1813     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1814     testl(rdx, rdx);
1815     jcc(Assembler::zero, L);
1816     call_VM(noreg, CAST_FROM_FN_PTR(address,
1817                                     InterpreterRuntime::post_method_entry));
1818     bind(L);
1819   }
1820 
1821   if (DTraceMethodProbes) {
1822     get_method(rarg);
1823     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1824                  rthread, rarg);
1825   }
1826 
1827   // RedefineClasses() tracing support for obsolete method entry
1828   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1829     get_method(rarg);
1830     call_VM_leaf(
1831       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1832       rthread, rarg);
1833   }
1834 }
1835 
1836 
1837 void InterpreterMacroAssembler::notify_method_exit(
1838     TosState state, NotifyMethodExitMode mode) {
1839   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1840   // track stack depth.  If it is possible to enter interp_only_mode we add
1841   // the code to check if the event should be sent.
1842   Register rthread = r15_thread;
1843   Register rarg = c_rarg1;
1844   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1845     Label L;
1846     // Note: frame::interpreter_frame_result has a dependency on how the
1847     // method result is saved across the call to post_method_exit. If this
1848     // is changed then the interpreter_frame_result implementation will
1849     // need to be updated too.
1850 
1851     // template interpreter will leave the result on the top of the stack.
1852     push(state);
1853     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1854     testl(rdx, rdx);
1855     jcc(Assembler::zero, L);
1856     call_VM(noreg,
1857             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1858     bind(L);
1859     pop(state);
1860   }
1861 
1862   if (DTraceMethodProbes) {
1863     push(state);
1864     get_method(rarg);
1865     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1866                  rthread, rarg);
1867     pop(state);
1868   }
1869 }
1870 
1871 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1872   // Get index out of bytecode pointer
1873   get_cache_index_at_bcp(index, 1, sizeof(u4));
1874   // Get address of invokedynamic array
1875   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1876   movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1877   if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
1878     shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
1879   } else {
1880     imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1881   }
1882   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
1883 }
1884 
1885 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1886   // Get index out of bytecode pointer
1887   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1888   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1889 
1890   movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
1891   // Take shortcut if the size is a power of 2
1892   if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1893     shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1894   } else {
1895     imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1896   }
1897   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
1898 }
1899 
1900 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1901   // Get index out of bytecode pointer
1902   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1903   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1904 
1905   movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
1906   imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1907   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
1908 }